Bug 9253 – Review Phobos algorithms and make them transient-safe where possible

Status
REOPENED
Severity
normal
Priority
P3
Component
phobos
Product
D
Version
D2
Platform
All
OS
All
Creation time
2013-01-01T12:09:55Z
Last change time
2024-12-01T16:16:03Z
Assigned to
No Owner
Creator
hsteoh
Moved to GitHub: phobos#9946 →

Comments

Comment #0 by hsteoh — 2013-01-01T12:09:55Z
This bug is to have a central place to keep the list of Phobos algorithms found to be transient-incompatible but could potentially be made transient-compatible, so that the list doesn't get lost in the dust of forum history. - std.algorithm.reduce (when no seed is given) - std.algorithm.joiner (both variants have been fixed in git HEAD) - std.algorithm.group - std.algorithm.minCount - std.algorithm.minPos (takes forward range; should use .save) - std.algorithm.Levenshtein (takes forward range; should use .save) - std.algorithm.makeIndex (takes forward range; should use .save) - std.algorithm.splitter (takes slices without checking for isSlicable) - std.algorithm.topNCopy - std.algorithm.NWayUnion - std.array.array (probably not fixable) - std.array.insertInPlace (probably not fixable) - std.array.join (copies input range; may not be fixable) - std.stdio.writeln & friends (need more testing, there are some deep bits that fail with transient ranges) While the whole transience issue hasn't been decided yet, Andrei has agreed that those algorithms that *can* be made transience-compatible, should be. The fate of the rest will be determined when this issue has been decided on.
Comment #1 by monarchdodra — 2013-01-01T13:05:23Z
(In reply to comment #0) > - std.algorithm.splitter (takes slices without checking for isSlicable) For the record, I'm on splitter. I had a pull ready, but closed it for further improvements.
Comment #2 by monarchdodra — 2013-01-14T13:30:18Z
(In reply to comment #0) > This bug is to have a central place to keep the list of Phobos algorithms found > to be transient-incompatible but could potentially be made > transient-compatible, so that the list doesn't get lost in the dust of forum > history. > > - std.algorithm.reduce (when no seed is given) > - std.algorithm.joiner (both variants have been fixed in git HEAD) > - std.algorithm.group > - std.algorithm.minCount > - std.algorithm.minPos (takes forward range; should use .save) > - std.algorithm.Levenshtein (takes forward range; should use .save) > - std.algorithm.makeIndex (takes forward range; should use .save) > - std.algorithm.splitter (takes slices without checking for isSlicable) > - std.algorithm.topNCopy > - std.algorithm.NWayUnion > - std.array.array (probably not fixable) > - std.array.insertInPlace (probably not fixable) > - std.array.join (copies input range; may not be fixable) > - std.stdio.writeln & friends (need more testing, there are some deep bits that > fail with transient ranges) > > While the whole transience issue hasn't been decided yet, Andrei has agreed > that those algorithms that *can* be made transience-compatible, should be. The > fate of the rest will be determined when this issue has been decided on. I just fixed minPos to use safe, and it should now be transient safe. No unittest though (yet) to prevent future breakage. I'm fixing minCount: It will be transient safe for forward ranges. Input ranges will the thoroughly unsafe though, with no possibility of workaround.
Comment #3 by hsteoh — 2013-01-14T21:44:56Z
Yeah, some algorithms will have to be transient-unsafe, because it will either introduce unacceptable overhead, or it's plain impossible due to the nature of the algorithm. These cases will just have to be left as-is.
Comment #4 by b2.temp — 2017-08-26T17:39:08Z
It looks like a failed initiative, not maintained since > 4 years. Since summer 2016 and the initiative to put annotations on all the unittest it's easier to locate the candidates.
Comment #5 by hsteoh — 2017-08-31T21:13:47Z
There must be some misunderstanding here. What has annotations got to do with transient ranges? Transient ranges, as referred to in this bug, are ranges where .front may mutate once .popFront is called, thereby making it invalid for code to cache .front by assigning to a local variable and referring to the variable later after .popFront is called. AFAIK there are no annotations that can be used for this. Many algorithms that currently break with transient ranges actually *can* be re-implemented in a way that doesn't break, and without undue overhead. Tracking these algorithms is the purpose of this issue.
Comment #6 by robert.schadek — 2024-12-01T16:16:03Z
THIS ISSUE HAS BEEN MOVED TO GITHUB https://github.com/dlang/phobos/issues/9946 DO NOT COMMENT HERE ANYMORE, NOBODY WILL SEE IT, THIS ISSUE HAS BEEN MOVED TO GITHUB